First class SQL support?

[JacobFV]

• add utilities to simplify applying tensor code on SQL objects: for example, using tensorcode to select rows in the database or define new tables
• since there are so many flavors of SQL, as well as other databases, do we really want to do this? Or would it be better to have users write their own extensions?

[JacobFV]

This might look like

with SQLiteModel(...):
   records = tc.get('uses in Dallas')
   records = tc.update('tally up the total runs for each user', records)
   tc.put('save the total runs', records)

And the class hierarchy might look like:

SQLiteModel > SQLModel > LLMBasedModel > BaseModel

[JacobFV]

However the above approach doesn't feel correct, since we might equally wish to use an NN-based model instead of an LLM-based one.

Perhaps, the SQL layer should be an wrapper around the model layer. Something like:

with SQLiteModel(..., model=tc.CURRENT_MODEL):
   ...